A Long COVID Trial Was Retracted. Half of It Still Stands

On October 6, 2026, Annals of Internal Medicine retracted a randomized trial of two cheap, widely available drugs for fatigue in long COVID, and on the same day published a corrected version of the same study. The trial, called REVIVE, had first appeared on March 31, 2026. The authors themselves asked for the retraction. The episode is a clean example of how a correction should work, and of why a single headline about a trial can outlive the data behind it.

What the trial tested

REVIVE was an adaptive randomized trial, a design that allows changes while the study runs, such as dropping an arm that is not working. It compared fluvoxamine, an antidepressant, and metformin, a diabetes drug, against placebo for fatigue in people with long COVID. Both drugs had been studied before in acute COVID, which made them natural candidates.

Adaptive designs are efficient, but they put a heavy load on reporting. When an arm changes dose halfway or stops early, the people in it no longer form one group treated one way, and the analysis has to say so. That is exactly the step that failed here.

What went wrong

The retraction notice, signed on behalf of all authors, is short and specific. During the trial, the independent data and safety monitoring committee, the DSMC, recommended two changes to the metformin arm. First, a dose reduction, from 750 mg twice daily to 500 mg twice daily, through a protocol amendment. Later, after a prespecified interim analysis, a stop for futility, meaning the arm was unlikely to show a benefit if it continued.

Both changes were carried out in the trial, and the trial data reflect them. But they were not carried through into the analysis and reporting of the metformin results in the published article. In the authors' words, those results did not accurately represent how the trial was conducted. The original paper presented the metformin arm as if it had run as first planned.

Six months passed between the first publication, on March 31, and the retraction. During that time the original metformin result was available to readers, reporters and reviewers as if it were final. Annals also retracted and republished the plain language summary for patients that accompanied the paper, so the lay version now matches the corrected science.

What changed in the new version

The republished article reports metformin by dose cohort, 750 mg and 500 mg twice daily, describes the DSMC's role in the Methods and Discussion, and calls the metformin findings inconclusive rather than definitive. The authors write that the fluvoxamine efficacy findings are unaffected and remain valid, and that the new version went through additional editorial and methodological review.

There is a small oddity worth noting for anyone who reads the source. The editors' note on the notice says the results of the fluoxetine arm are valid. The trial tested fluvoxamine, a different antidepressant, and the authors' own text and the titles say fluvoxamine. It reads as a slip in the note, not a change in the finding.

Why this is handled the right way

Nothing in the notice suggests that participants were put at risk by the error. The dose reduction and the futility stop were decisions of the independent committee during the trial, which is the job such committees exist to do; what failed was the paper's account of them.

Retractions often signal misconduct. This one does not. The data were collected as the trial was actually run; the error was in analysis and reporting, and the authors flagged it. Annals chose retraction and republication, a route that removes the flawed version from the record while keeping the valid work available, rather than a quiet correction that readers of the first paper might never see. The original citation, volume 179, pages 621 to 628, now points to a retracted article.

The practical lesson is for readers of trial news. When a trial is adaptive, check how many arms were changed or stopped along the way, and whether the paper analyzes them as they actually ran. That is one of the questions in our five minute guide to reading a health study. This article describes a research record, not treatment guidance: anyone with long COVID weighing either drug should discuss it with their own clinician.

I think the most important number here is the one that disappeared. A clear metformin result is now an open question, and the coverage of the March paper will not update itself. The fluvoxamine finding survives the correction, which says something good about the rest of the trial. But the cleanest way to see the change is to read the two versions side by side, and most people who saw the first headline never will.